Testing for sample selection bias due to location effects in work history data collected retrospectively from a geographically based sample.

Crouchley, Rob (1998) Testing for sample selection bias due to location effects in work history data collected retrospectively from a geographically based sample. Environment and Planning A, 30 (12). pp. 2195-2210. ISSN 0308-518X

Abstract

A three-state competing-risk labour-market model with a submodelfor the state on entry into the labour market and a control for individual-level heterogeneity is constructed. A test for sample-selection bias due to location effects in the labour-market model is formulated and applied to the retrospective work-history data from the Social Class and Economic Life Initiative (SCELI). The test suggests that location and labour-market behaviour are interdependent outcomes. As the SCELI has sampled data from a very restricted subset of all possible locations in Britain the model loses external validity, that is, the results cannot be generalised to individuals who live outside the selected areas at the sample date. The model also loses internal validity, because selection on location will have induced a correlation between the random components and many of the covariates in the model, implying that the estimated parameters are misleading indicators of the true covariate effects. These conclusions must therefore raise questions about the validity of any substantive research related to labour-market outcomes which uses the SCELI data.